Small edits, large models: How Wikipedia advocacy shapes LLM values
Small edits, large models: How Wikipedia advocacy shapes LLM values
小编辑，大模型：维基百科倡导如何塑造LLM价值观
Authors: Jasmine Brazilek, Maria Navas, Alexa Gnauck
作者： Jasmine Brazilek、Maria Navas、Alexa Gnauck
Can a small group of volunteers shape how AI systems discuss animal welfare, just by editing Wikipedia? We show that they can. Wikipedia appears in nearly every major language model training dataset and is weighted more heavily than web-crawled text. The Pro-Animal Wikipedians (PAW), a group of advocates who add sourced animal welfare content to relevant articles, have made 125 edits across 115 pages. Using gradient-based data attribution (Bergson; MAGIC), we traced how these edits influence language model behavior. TrackStar retrieval attribution on Llama 3.1 8B found that PAW-edited sections made up 68 percent of the highest-attributed documents for animal welfare queries (p < 0.0001) but only 52 percent for unrelated queries about the same companies (p = 0.53): the model links PAW content specifically to animal welfare topics, not to the entities in general. MAGIC counterfactual influence estimation on Llama-3.2-1B, run across five random training-order seeds, gave the same picture even more sharply: in every seed, the top-10 most influential documents on animal welfare queries were all PAW edits (10 of 10, 5 of 5 seeds), while on general queries the same top-10 sat at chance (4 to 6 of 10). Mean PAW influence exceeded mean control influence on animal welfare queries with p < 0.0001 in every seed, an effect 6 to 30 times larger than on general queries. Leave-subset-out validation gave Spearman rho = 1.00 for all 10 runs. When we fine-tuned separate models on PAW content versus control content, each model performed better specifically on the type of text it was trained on: the PAW-trained model cut perplexity on animal welfare text from 12.4 to 8.4, while the control-trained model cut perplexity on control text from 16.1 to 11.4. A small, coordinated Wikipedia editing campaign therefore measurably shapes how language models handle the topics those edits address.
仅仅通过编辑维基百科，一小群志愿者能否塑造AI系统讨论动物福利的方式？我们证明他们确实可以。维基百科几乎出现在所有主要语言模型的训练数据集中，并且比网络爬取文本的权重更高。亲动物维基百科编辑者（PAW）——一群在相关文章中添加强调来源的动物福利内容的倡导者——在115个页面上进行了125次编辑。我们使用基于梯度的数据归因方法（Bergson; MAGIC，基于梯度的数据归因）追踪了这些编辑如何影响语言模型的行为。在Llama 3.1 8B上进行的TrackStar检索归因（TrackStar retrieval attribution）发现，对于动物福利查询，PAW编辑的章节占最高归因文档的68%（p < 0.0001），但对于针对同一公司的不相关查询，这一比例仅为52%（p = 0.53）：模型将PAW内容与动物福利主题特异地关联起来，而非与实体本身相关联。在Llama-3.2-1B上跨五个随机训练顺序种子的MAGIC反事实影响估计（MAGIC counterfactual influence estimation）更清晰地揭示了同样的情况：在每个种子中，对动物福利查询影响最大的前10个文档全部是PAW编辑的内容（10/10，5/5种子），而对于一般查询，同样的前10个文档则处于随机水平（4到6/10）。在每个种子中，PAW的平均影响在动物福利查询上显著超过对照平均影响（p < 0.0001），其效应大小是对一般查询的6到30倍。留子集验证（Leave-subset-out validation）在所有10次运行中给出了Spearman rho = 1.00。当我们在PAW内容与对照内容上分别微调模型时，每个模型在其训练文本类型上的表现更好：PAW训练模型将动物福利文本的困惑度从12.4降至8.4，而对照训练模型将对照文本的困惑度从16.1降至11.4。因此，一场小型、协调的维基百科编辑活动确实可衡量地塑造了语言模型处理这些编辑所涉及主题的方式。